CN111310656A - Single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis - Google Patents

Single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis Download PDF

Info

Publication number
CN111310656A
CN111310656A CN202010091776.4A CN202010091776A CN111310656A CN 111310656 A CN111310656 A CN 111310656A CN 202010091776 A CN202010091776 A CN 202010091776A CN 111310656 A CN111310656 A CN 111310656A
Authority
CN
China
Prior art keywords
data
principal component
component analysis
analysis
tensor
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010091776.4A
Other languages
Chinese (zh)
Inventor
付荣荣
杨阳
于宝
王世伟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yanshan University
Original Assignee
Yanshan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Yanshan University filed Critical Yanshan University
Priority to CN202010091776.4A priority Critical patent/CN111310656A/en
Publication of CN111310656A publication Critical patent/CN111310656A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/15—Biometric patterns based on physiological signals, e.g. heartbeat, blood flow

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Signal Processing (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

The invention provides a single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis, which can find a projection matrix from a time domain, a frequency domain and a space domain respectively to project 3-dimensional EEG tensor data so as to realize the dimension reduction of original EEG tensor data, and then combines a linear classification method to classify. Compared with the traditional principal component analysis method, the multilinear principal component analysis method provided by the invention directly reduces the dimensions from multiple dimensions in the multidimensional tensor, reserves the spatial structure information of the signals, and expands the signals into a one-dimensional vector form for classification after dimension reduction, so that compared with the traditional principal component analysis-based method, the method provided by the invention reserves the spatial characteristics of EEG signals; compared with EEG time domain analysis, frequency domain analysis, time frequency analysis or time-space analysis, the EEG signal multi-modal analysis method based on the time domain, frequency domain and space domain can extract more comprehensive characteristics, and the brain electrical identification effect is still high under the condition of a small sample.

Description

Single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis
Technical Field
The invention relates to the field of biological signal processing, in particular to a single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis.
Background
An electroencephalogram (EEG) is a signal generated by brain neuron activity, which contains rich brain state information, and needs to be effectively decoded to realize a brain-computer interface (BCI), and the decoding process includes feature extraction and pattern classification of the EEG signal. In recent years, many international research groups have invested much effort in the feature extraction method of single motor imagery electroencephalography (MI-EEG). The method for directly extracting features from the time domain is the earliest developed method because of strong intuition and clear physical significance. However, because the waveform of the electroencephalogram signal is too complex, no particularly effective EEG time domain analysis method exists at present; because many main characteristics of the EEG signal are reflected on a frequency domain, and power spectrum estimation is an important means of frequency domain analysis, the spectrum analysis technology occupies a particularly important position in EEG signal processing, but the power spectrum estimation cannot reflect the time-varying property of an EEG frequency spectrum, so that time-varying information can be lost from the power spectrum estimation of the frequency domain for a time-varying non-stationary process such as EEG; the time-frequency analysis technology of signals is different from the prior pure time domain or frequency domain analysis, is a technology for analyzing signals in time and frequency domains at the same time, and is mainly divided into two types of linear change and nonlinear transformation. The most widely used method at present is the wavelet transform theory. The wavelet analysis uses a short window at high frequency and uses a wide window at low frequency, so that the ideas of relative bandwidth frequency analysis and adaptive variable resolution analysis are fully embodied, and a possible path is provided for real-time analysis of signals. At present, time-frequency analysis research of electroencephalogram signals has obtained a plurality of valuable research results. However, the wavelet transformation effect depends heavily on the selection of center frequency and bandwidth in continuous wavelet transformation, at present, the selection of the parameters usually depends on experience or experiment, and the effect of the electroencephalogram signals with large individual difference is not stable enough; the spatial distribution of the brain electricity on the scalp is considered, and the time-space analysis method for fusion analysis of the time and space information is beneficial to revealing and enhancing implicit characteristics in the multi-lead brain electricity signal. The time-space analysis method can provide more information for people, is an important research direction in EEG signal analysis, but ignores frequency domain information contained in EEG signals.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a signal analysis method capable of obtaining more comprehensive EEG signal characteristics. Aiming at the technical problems, the invention provides a single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis, the method can find a projection matrix from a time domain, a frequency domain and a space domain respectively to project 3-dimensional EEG tensor data so as to realize the dimension reduction of original EEG tensor data, and then the classification is carried out by combining a linear classification method.
The method comprises the following specific steps:
step 1, establishing third-order EEG tensor data of multiple experiments by using a wavelet analysis method, and randomly dividing the third-order EEG tensor data into a training set and a test set, wherein the method comprises the following specific steps:
step 11, when a subject imagines the movement of the left hand and the right hand according to the prompt, acquiring electroencephalogram data of the subject, and sequentially intercepting the acquired electroencephalogram data during each motor imagery according to the prompt time point to form a single motor imagery electroencephalogram data matrix comprising two dimensions of time and space, wherein the space represents different acquisition channels;
step 12, performing band-pass filtering on the electroencephalogram data;
step 13, extracting frequency domain information contained in the electroencephalogram signals by adopting complex Morlet wavelet transformation, and constructing the extracted frequency domain, time domain and space domain data into third-order tensor
Figure BDA0002383315390000021
Wherein c represents a channel, f represents a frequency, and t represents a time;
step 14, randomly dividing the third order tensor data into training sets
Figure BDA0002383315390000022
And test set
Figure BDA0002383315390000023
Selecting an optimal classification model by adopting a cross validation mode;
step 2, training the test set by utilizing a multi-linear principal component analysis method to obtain a multi-modal dimensionality reduction projection matrix, and projecting the training set to obtain dimensionality reduction training set data, wherein the method comprises the following specific steps:
step 21, for the tensor sample
Figure BDA0002383315390000031
The centralization treatment is carried out, and the treatment is carried out,
Figure BDA0002383315390000032
in the formula
Figure BDA0002383315390000033
Figure BDA0002383315390000034
In order to obtain a centered sample,
Figure BDA0002383315390000035
in the form of an original sample, the sample is,
Figure BDA0002383315390000036
is the sample mean value, M is the number of samples;
step 22, calculating an initial covariance matrix Xm(n)Xm(n) TIn the formula Xm(n)Is Xmn-mode expanded matrix, performing characteristic decomposition on the initial covariance matrix, and forming a projection matrix U by using eigenvectors corresponding to the largest d' eigenvalues(n)(n is 1,2,3), obtaining an initialized dimension reduction projection matrix;
step 23, performing local optimization on the initialized dimension-reduced projection matrix obtained in the step 22, wherein the local optimization includes the following specific steps;
step (ii) of231. Performing projection
Figure BDA0002383315390000037
In the formula, the left subscripts 1,2 and 3 represent the product of the 1 mode, the 2 mode and the 3 mode;
step 232, calculate the total divergence
Figure BDA0002383315390000038
In the formula
Figure BDA0002383315390000039
Is a tensor norm;
step 233, for n being 1,2, and 3, computing eigenvalues of covariance matrix after tensor n mode expansion after projection, and combining eigenvectors corresponding to d' eigenvalues to form a new projection matrix U(n)(n-1, 2,3) updating with the new projection matrix
Figure BDA00023833153900000310
And calculate what is new
Figure BDA00023833153900000311
Step 234, judge
Figure BDA00023833153900000312
If yes, k is the optimization iteration number, η is the self-defined threshold value, if yes
Figure BDA00023833153900000313
Obtaining a final projection matrix1U(1),2U(2),3U(3)Calculating
Figure BDA00023833153900000314
Obtaining training set data subjected to multi-linear principal component analysis and dimensionality reduction; otherwise steps 231, 232 and 233 are repeated;
and 3, training a classifier after feature selection is carried out by using the dimensionality reduced training set data to obtain an optimal classification model, and the method comprises the following specific steps of:
step 31, expanding the training set data in the three-order tensor form after dimension reduction into a one-dimensional array, and calculating the in-class dispersion of each characteristic component
Figure BDA0002383315390000041
Degree of inter-class dispersion
Figure BDA0002383315390000042
And the ratio of the two
Figure BDA0002383315390000043
Wherein K is 1,2, …, K, K is the number of categories; mkThe number of samples of each type; vmIs a k type sample;
Figure BDA0002383315390000044
mean value of each type of sample;
Figure BDA0002383315390000045
is the average value of the whole samples;
step 32, sorting the characteristic components according to the size of J, only reserving the largest front D group of characteristic components, wherein D is a self-defined characteristic number, and the optimal characteristic number can be found by multiple attempts;
step 33, training the classifier by using the obtained data to obtain a classification projection matrix Uclassify；
And 4, checking the final classification performance by using the data of the test set after dimensionality reduction to obtain classification accuracy, wherein the method comprises the following specific steps of:
step 41, subjecting the test set data to multi-linear principal component analysis and dimension reduction
Figure BDA0002383315390000046
Expanding the same into a one-dimensional array selection characteristic;
step 42, passing through a classification matrix UclassifyAnd projecting to obtain categories and finally obtaining the classification accuracy.
Preferably, the electroencephalogram data with the duration of 1 second is sequentially intercepted from the acquired electroencephalogram data according to the prompted time point in the step 11.
Preferably, in the step 12, the passband range of the band-pass filtering is 8 to 13 Hz.
Preferably, in step 13, the bandwidth parameter of the complex Morlet wavelet is 1Hz, and the center frequency is 2 Hz.
Preferably, in the step 14, the third-order tensor data is randomly divided into nine-tenth training sets
Figure BDA0002383315390000047
And one tenth of the test set
Figure BDA0002383315390000048
And selecting the optimal classification model by adopting a 10-fold cross validation mode.
Preferably, in the step 234, the threshold η is 10-6。
Preferably, in the step 32, the customized feature number D is 35.
Compared with the prior art, the invention has the following beneficial effects:
in the traditional principal component analysis method, a multi-dimensional signal is directly unfolded into a one-dimensional vector form, so that the dimension reduction process is always carried out in the one-dimensional vector, and the spatial structure of an electroencephalogram signal, namely electroencephalogram channel information acquired by the signal, is lost. The single motor imagery electroencephalogram signal identification method based on the multi-linear principal component analysis directly reduces dimensions from multiple dimensions in a multi-dimensional tensor, reserves the spatial structure information of the signals, and expands the signals into a one-dimensional vector form for classification after dimension reduction, so that compared with the traditional method based on the principal component analysis, the method provided by the invention reserves the spatial characteristics of an EEG signal;
compared with EEG time domain analysis, frequency domain analysis, time frequency analysis or time-space analysis, the EEG signal multi-modal analysis method based on the time domain, frequency domain and space domain can extract more comprehensive characteristics, and the brain electrical identification effect is still high under the condition of a small sample.
Drawings
FIG. 1 is a general flow chart of a single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis, which is provided by the embodiment of the invention;
FIG. 2 is a schematic diagram of specific operation steps of a single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis according to an embodiment of the present invention; and
FIG. 3 is a schematic diagram of the identification accuracy of the single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis in the embodiment of the present invention.
Detailed Description
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
The general flow chart of the single motor imagery electroencephalogram signal identification method based on the multi-linear principal component analysis, which is provided by the embodiment of the invention, is shown in figure 1, and the method comprises the following steps:
step 1, establishing third-order EEG tensor data of multiple experiments by using a wavelet analysis method, and randomly dividing the third-order EEG tensor data into a training set and a testing set, wherein the method comprises the following specific steps:
step 11, when a subject imagines the movement of the left hand and the right hand according to the prompt, a high-precision mobile brain wave test instrument EMOTIV EPOC +14 is adopted to collect the electroencephalogram data of the subject, the collected continuous electroencephalogram signals are intercepted into the electroencephalogram data of each channel of each subject during each motor imagery according to the prompt time point, finally, a time-space two-dimensional single motor imagery electroencephalogram data matrix of each subject is formed, the space is different collection channels, and the time length is 1 second from the prompt time point;
step 12, performing 8-13 Hz band-pass filtering on the signals;
step 13, extracting frequency domain information contained in the electroencephalogram signals by adopting complex Morlet wavelet transformation with a bandwidth parameter of 1Hz and a wavelet center frequency of 2Hz, and constructing the extracted frequency domain, time domain and space domain data into third-order tensor
Figure BDA0002383315390000061
Wherein c represents a channel, f represents a frequency, and t represents a time;
step 14, because the sample size is small, the embodiment randomly divides the data into nine-tenth training sets
Figure BDA0002383315390000062
And one tenth of the test set
Figure BDA0002383315390000063
Selecting an optimal classification model by adopting a 10-fold cross validation mode;
step 2, training the test set by using a multi-linear principal component analysis method to obtain a multi-modal dimensionality reduction projection matrix, and projecting the training set, wherein as shown in fig. 2, the method comprises the following specific steps:
step 21, for the original tensor sample
Figure BDA0002383315390000064
The centralization treatment is carried out, and the treatment is carried out,
Figure BDA0002383315390000065
wherein
Figure BDA0002383315390000066
Figure BDA0002383315390000067
In order to obtain a centered sample,
Figure BDA0002383315390000068
in the form of an original sample, the sample is,
Figure BDA0002383315390000069
is the sample mean value, M is the number of samples;
step 22, calculating an initial covariance matrix Xm(n)Xm(n) T，Xm(n)Is Xmn-mode expanded matrix, performing characteristic decomposition, and forming projection matrix U by using eigenvectors corresponding to the largest d' eigenvalues(n)(n is 1,2,3), obtaining an initialized dimension reduction projection matrix;
step 23, performing local optimization on the initialized dimension reduction projection matrix obtained in the step 22, wherein the method comprises the following specific steps;
step 231, projecting
Figure BDA00023833153900000610
In the formula, the left subscripts 1,2 and 3 represent the product of the 1 mode, the 2 mode and the 3 mode;
step 232, calculate the total divergence
Figure BDA0002383315390000071
In the formula
Figure BDA0002383315390000072
Is a tensor norm;
step 233, computing eigen decomposition of the covariance matrix after n-mode expansion of the tensor after projection as 1,2 and 3, and forming a new projection matrix U by using eigenvectors corresponding to d' eigenvalues(n)(n is 1,2,3), updated with the new projection matrix
Figure BDA0002383315390000073
And calculate what is new
Figure BDA0002383315390000074
Step 234, judge
Figure BDA0002383315390000075
If yes, k is the optimization iteration number, η is the self-defined threshold value, if yes
Figure BDA0002383315390000076
Obtaining a final projection matrix1U(1),2U(2),3U(3)Calculating
Figure BDA0002383315390000077
Obtaining training set data subjected to multi-linear principal component analysis and dimensionality reduction; otherwise steps 231, 232 and 233 are repeated;
and 3, training a classifier after feature selection is carried out by using the training set data subjected to dimensionality reduction to obtain an optimal classification model, and the method comprises the following specific steps of:
step 31, the training set data after dimensionality reduction is in a multi-group third-order tensor form, the training set data is firstly unfolded into a one-dimensional array, and the in-class dispersion of each characteristic component is calculated
Figure BDA0002383315390000078
Degree of inter-class dispersion
Figure BDA0002383315390000079
And the ratio of the two
Figure BDA00023833153900000710
Wherein K is 1,2, …, K, K is the number of categories; mkThe number of samples of each type; vmIs a k type sample;
Figure BDA00023833153900000711
mean value of each type of sample;
Figure BDA00023833153900000712
is the average value of the whole samples;
step 32, sorting each feature component according to the size of J, only reserving the largest front D group feature component, wherein D is a self-defined feature number, and the best feature number can be found by multiple attempts;
step 33, training the classifier by using the obtained data to obtain a classification projection matrix Uclassify。
And 4, checking the final classification performance by using the data of the test set after dimensionality reduction to obtain classification accuracy, wherein the method comprises the following specific steps of:
step 41, subjecting the test set data to multi-linear principal component analysis and dimension reduction
Figure BDA0002383315390000081
Expanding the same into a one-dimensional array selection characteristic;
step 42, passing through a classification matrix UclassifyAnd projecting to obtain categories and finally obtaining the classification accuracy. Recognition accuracy of the present embodimentAs shown in fig. 3, the average of the multiple sets of accuracy rates is taken to obtain the accuracy rate (ACC) of 94.4% in the present scheme.
The invention designs an EEG tensor multi-modal analysis method, which can read task-related time domain, space and frequency domain discrimination projection modes from complex EEG signals, and can obtain more comprehensive characteristics compared with simple time domain, frequency domain analysis or time frequency analysis and time-space analysis, thereby obtaining better mode identification effect, and the average accuracy of multiple experiments can reach 94.4%.
The above-mentioned embodiments are merely illustrative of the preferred embodiments of the present invention, and do not limit the scope of the present invention, and various modifications and improvements made to the technical solution of the present invention by those skilled in the art without departing from the spirit of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims (7)

1. A single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis is characterized by comprising the following specific steps:
step 1, establishing third-order EEG tensor data of multiple experiments by using a wavelet analysis method, and randomly dividing the third-order EEG tensor data into a training set and a test set, wherein the method comprises the following specific steps:
step 11, when a subject imagines the movement of the left hand and the right hand according to the prompt, acquiring electroencephalogram data of the subject, and sequentially intercepting the acquired electroencephalogram data during each motor imagery according to the prompt time point to form a single motor imagery electroencephalogram data matrix comprising two dimensions of time and space, wherein the space represents different acquisition channels;
step 12, performing band-pass filtering on the electroencephalogram data;
step 13, extracting frequency domain information contained in the electroencephalogram signals by adopting complex Morlet wavelet transformation, and constructing the extracted frequency domain, time domain and space domain data into third-order tensor
Figure FDA0002383315380000011
Wherein c represents a channel, f represents a frequency, and t represents a time;
step 14, randomly dividing the third order tensor data into training sets
Figure FDA0002383315380000012
And test set
Figure FDA0002383315380000013
Selecting an optimal classification model by adopting a cross validation mode;
step 2, training the test set by utilizing a multi-linear principal component analysis method to obtain a multi-modal dimensionality reduction projection matrix, and projecting the training set to obtain dimensionality reduction training set data, wherein the method comprises the following specific steps:
step 21, for the tensor sample
Figure FDA0002383315380000014
The centralization treatment is carried out, and the treatment is carried out,
Figure FDA0002383315380000015
in the formula
Figure FDA0002383315380000016
Figure FDA0002383315380000017
In order to obtain a centered sample,
Figure FDA0002383315380000018
in the form of an original sample, the sample is,
Figure FDA0002383315380000019
is the sample mean value, M is the number of samples;
step 22, calculating an initial covariance matrix Xm(n)Xm(n) TIn the formula Xm(n)Is Xmn-mode expanded matrix and characteristic decomposition is carried out on the initial covariance matrixAnd forming a projection matrix U by using eigenvectors corresponding to the largest d' eigenvalues(n)(n is 1,2,3), obtaining an initialized dimension reduction projection matrix;
step 23, performing local optimization on the initialized dimension-reduced projection matrix obtained in the step 22, wherein the local optimization includes the following specific steps;
step 231, projecting
Figure FDA0002383315380000021
In the formula, the left subscripts 1,2 and 3 represent the product of the 1 mode, the 2 mode and the 3 mode;
step 232, calculate the total divergence
Figure FDA0002383315380000022
In the formula
Figure FDA0002383315380000023
Is a tensor norm;
step 233, for n being 1,2, and 3, computing eigenvalues of covariance matrix after tensor n mode expansion after projection, and combining eigenvectors corresponding to d' eigenvalues to form a new projection matrix U(n)(n-1, 2,3) updating with the new projection matrix
Figure FDA0002383315380000024
And calculate what is new
Figure FDA0002383315380000025
Step 234, judge
Figure FDA0002383315380000026
If yes, k is the optimization iteration number, η is the self-defined threshold value, if yes
Figure FDA0002383315380000027
Obtaining a final projection matrix1U(1),2U(2),3U(3)Calculating
Figure FDA0002383315380000028
Obtaining training set data subjected to multi-linear principal component analysis and dimensionality reduction; otherwise steps 231, 232 and 233 are repeated;
and 3, training a classifier after feature selection is carried out by using the dimensionality reduced training set data to obtain an optimal classification model, and the method comprises the following specific steps of:
step 31, expanding the training set data in the three-order tensor form after dimension reduction into a one-dimensional array, and calculating the in-class dispersion of each characteristic component
Figure FDA0002383315380000029
Degree of inter-class dispersion
Figure FDA00023833153800000210
And the ratio of the two
Figure FDA00023833153800000211
Wherein K is 1,2, …, K, K is the number of categories; mkThe number of samples of each type; vmIs a k type sample;
Figure FDA00023833153800000212
mean value of each type of sample;
Figure FDA00023833153800000213
is the average value of the whole samples;
step 32, sorting the characteristic components according to the size of J, only reserving the largest front D group of characteristic components, wherein D is a self-defined characteristic number, and the optimal characteristic number can be found by multiple attempts;
step 33, training the classifier by using the obtained data to obtain a classification projection matrix Uclassify；
And 4, checking the final classification performance by using the data of the test set after dimensionality reduction to obtain classification accuracy, wherein the method comprises the following specific steps of:
step 41, test setThe data is subjected to multi-linear principal component analysis for dimensionality reduction
Figure FDA0002383315380000031
Expanding the same into a one-dimensional array selection characteristic;
step 42, passing through a classification matrix UclassifyAnd projecting to obtain categories and finally obtaining the classification accuracy.
2. The single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis according to claim 1, wherein electroencephalogram data with a duration of 1 second are sequentially intercepted from the acquired electroencephalogram data according to a prompted time point in step 11.
3. The single motor imagery electroencephalogram signal identification method based on the multi-linear principal component analysis of claim 1, wherein in the step 12, the pass band range of the band-pass filtering is 8-13 Hz.
4. The single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis according to claim 1, wherein in step 13, the complex Morlet wavelet has a bandwidth parameter of 1Hz and a center frequency of 2 Hz.
5. The method for single motor imagery electroencephalogram signal identification based on multi-linear principal component analysis of claim 1, wherein in step 14, the third order tensor data are randomly divided into nine-tenth training sets
Figure FDA0002383315380000032
And one tenth of the test set
Figure FDA0002383315380000033
And selecting the optimal classification model by adopting a 10-fold cross validation mode.
6. According to claim1, the single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis is characterized in that in the step 234, the threshold η is 10-6。
7. The method for single motor imagery electroencephalogram signal identification based on multi-linear principal component analysis of claim 1, wherein in the step 32, the customized feature number D is 35.
CN202010091776.4A 2020-02-13 2020-02-13 Single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis Pending CN111310656A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010091776.4A CN111310656A (en) 2020-02-13 2020-02-13 Single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010091776.4A CN111310656A (en) 2020-02-13 2020-02-13 Single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis

Publications (1)

Publication Number Publication Date
CN111310656A true CN111310656A (en) 2020-06-19

Family

ID=71156589

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010091776.4A Pending CN111310656A (en) 2020-02-13 2020-02-13 Single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis

Country Status (1)

Country Link
CN (1) CN111310656A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112036354A (en) * 2020-09-09 2020-12-04 东南大学 Natural action electroencephalogram recognition method based on Riemann geometry
CN113143295A (en) * 2021-04-23 2021-07-23 河北师范大学 Equipment control method and terminal based on motor imagery electroencephalogram signals
CN113158793A (en) * 2021-03-15 2021-07-23 东北电力大学 Multi-class motor imagery electroencephalogram signal identification method based on multi-feature fusion
CN113220120A (en) * 2021-04-27 2021-08-06 武汉理工大学 Self-adaptive motor imagery brain-computer interface training method fusing subjective and objective evaluation
CN113349795A (en) * 2021-06-15 2021-09-07 杭州电子科技大学 Depression electroencephalogram analysis method based on sparse low-rank tensor decomposition

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103116764A (en) * 2013-03-02 2013-05-22 西安电子科技大学 Brain cognitive state judgment method based on polyteny principal component analysis
CN104586387A (en) * 2015-01-19 2015-05-06 秦皇岛市惠斯安普医学***有限公司 Method for extracting and fusing time, frequency and space domain multi-parameter electroencephalogram characters
CN105956624A (en) * 2016-05-06 2016-09-21 东南大学 Motor imagery electroencephalogram classification method based on space-time-frequency optimization feature sparse representation
CN108960299A (en) * 2018-06-15 2018-12-07 东华大学 A kind of recognition methods of multiclass Mental imagery EEG signals
CN109472194A (en) * 2018-09-26 2019-03-15 重庆邮电大学 A kind of Mental imagery EEG signals characteristic recognition method based on CBLSTM algorithm model
CN109480834A (en) * 2018-12-20 2019-03-19 浙江大学 A kind of Method of EEG signals classification based on quick multiple dimension empirical mode decomposition

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103116764A (en) * 2013-03-02 2013-05-22 西安电子科技大学 Brain cognitive state judgment method based on polyteny principal component analysis
CN104586387A (en) * 2015-01-19 2015-05-06 秦皇岛市惠斯安普医学***有限公司 Method for extracting and fusing time, frequency and space domain multi-parameter electroencephalogram characters
CN105956624A (en) * 2016-05-06 2016-09-21 东南大学 Motor imagery electroencephalogram classification method based on space-time-frequency optimization feature sparse representation
CN108960299A (en) * 2018-06-15 2018-12-07 东华大学 A kind of recognition methods of multiclass Mental imagery EEG signals
CN109472194A (en) * 2018-09-26 2019-03-15 重庆邮电大学 A kind of Mental imagery EEG signals characteristic recognition method based on CBLSTM algorithm model
CN109480834A (en) * 2018-12-20 2019-03-19 浙江大学 A kind of Method of EEG signals classification based on quick multiple dimension empirical mode decomposition

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
王金甲 等: "脑机接口中多线性主成分分析的张量特征提取", 《生物医学工程学杂志》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112036354A (en) * 2020-09-09 2020-12-04 东南大学 Natural action electroencephalogram recognition method based on Riemann geometry
CN112036354B (en) * 2020-09-09 2022-04-29 东南大学 Natural action electroencephalogram recognition method based on Riemann geometry
CN113158793A (en) * 2021-03-15 2021-07-23 东北电力大学 Multi-class motor imagery electroencephalogram signal identification method based on multi-feature fusion
CN113158793B (en) * 2021-03-15 2023-04-07 东北电力大学 Multi-class motor imagery electroencephalogram signal identification method based on multi-feature fusion
CN113143295A (en) * 2021-04-23 2021-07-23 河北师范大学 Equipment control method and terminal based on motor imagery electroencephalogram signals
CN113220120A (en) * 2021-04-27 2021-08-06 武汉理工大学 Self-adaptive motor imagery brain-computer interface training method fusing subjective and objective evaluation
CN113220120B (en) * 2021-04-27 2022-08-30 武汉理工大学 Self-adaptive motor imagery brain-computer interface training method fusing subjective and objective evaluation
CN113349795A (en) * 2021-06-15 2021-09-07 杭州电子科技大学 Depression electroencephalogram analysis method based on sparse low-rank tensor decomposition
CN113349795B (en) * 2021-06-15 2022-04-08 杭州电子科技大学 Depression electroencephalogram analysis method based on sparse low-rank tensor decomposition

Similar Documents

Publication Publication Date Title
CN111310656A (en) Single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis
Lakshmi et al. Survey on EEG signal processing methods
Ince et al. Adapting subject specific motor imagery EEG patterns in space–time–frequency for a brain computer interface
CN110781945A (en) Electroencephalogram signal emotion recognition method and system integrating multiple features
Anderson et al. Geometric subspace methods and time-delay embedding for EEG artifact removal and classification
Darvishi et al. Brain-computer interface analysis using continuous wavelet transform and adaptive neuro-fuzzy classifier
CN104586387A (en) Method for extracting and fusing time, frequency and space domain multi-parameter electroencephalogram characters
Zamanian et al. A new feature extraction method to improve emotion detection using EEG signals
CN114533086B (en) Motor imagery brain electrolysis code method based on airspace characteristic time-frequency transformation
CN113158793A (en) Multi-class motor imagery electroencephalogram signal identification method based on multi-feature fusion
CN111184509A (en) Emotion-induced electroencephalogram signal classification method based on transfer entropy
CN112890827B (en) Electroencephalogram identification method and system based on graph convolution and gate control circulation unit
CN110399846A (en) A kind of gesture identification method based on multichannel electromyography signal correlation
CN112515685A (en) Multi-channel electroencephalogram signal channel selection method based on time-frequency co-fusion
CN109657646B (en) Method and device for representing and extracting features of physiological time series and storage medium
CN113536882B (en) Multi-class motor imagery electroencephalogram signal feature extraction and classification method
CN109009098B (en) Electroencephalogram signal feature identification method under motor imagery state
Jinliang et al. EEG emotion recognition based on granger causality and capsnet neural network
Taherisadr et al. EEG signal dimensionality reduction and classification using tensor decomposition and deep convolutional neural networks
CN115414051A (en) Emotion classification and recognition method of electroencephalogram signal self-adaptive window
CN106503733A (en) Based on the useful signal recognition methods that NA MEMD and GMM are clustered
Nakra et al. Feature Extraction and Dimensionality Reduction Techniques with Their Advantages and Disadvantages for EEG-Based BCI System: A Review.
Azami et al. Automatic signal segmentation based on singular spectrum analysis and imperialist competitive algorithm
CN114358090B (en) Motor imagery electroencephalogram signal classification method based on PSD and CSP
CN107423668A (en) Eeg signal classification System and method for based on wavelet transformation and sparse expression

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20200619